No – paydown runs in parallel, module by module, with your team spending review time, not implementation time.

Through the debt map's scoring – dependencies, duplication, complexity, and coverage gaps — re-scored after each paydown phase.

Yes, if agent output ships without review – this is the most common failure mode in 2026. Every agent-proposed change here goes through a human approval gate before merge specifically to prevent it.

Outdated dependencies, dead code, duplicated logic, missing tests, and undocumented modules are strong fits. Deep architectural decisions still need human-led design work.

Yes. Agents operating without documented conventions – naming, error handling, architectural boundaries – tend to generate code that compiles but violates decisions that become expensive to undo later. We codify this context (in a CLAUDE.md-style file) so agents work within your architecture instead of around it.

Core domain logic, security-critical components, and anything touching inter-service contracts require senior engineer sign-off regardless of AI involvement – these categories are defined explicitly before any agent-assisted work starts, not decided case by case.

Credentials, PII, and security-critical modules are excluded from agent access before work begins – this governance is defined upfront, not handled reactively after something's already been exposed.

No – it reallocates their time. Senior engineers set the architectural standards agents operate within and review the changes where judgment matters (domain logic, security, cross-service impact); they spend less time on scaffolding and routine refactoring, not less time on decisions.